The approval and subsequent withdrawal of widely prescribed unsafe drugs affects millions of lives annually. We have recently reported that a Bayesian network model can utilize preclinical, phase I and phase II data to predict phase III safety and efficacy with 78% accuracy. Our approach exceeds pharmaceutical industry performance. Our novel preliminary data demonstrate that identifying post-marketing safety issues independent of drug class is feasible using preclinical data only. Every drug has a unique set of preclinical dose versus primary effect curves and dose versus side effect curves. Our preliminary studies show that quantifiable features of preclinical dose versus primary effect curves predict post-approval safety withdrawal with impressive accuracy. Our objective is to build and distribute preclinical pharmacologic predictive models of post-approval clinical safety. Our specific aims are (1) to identify preclinical dose-effect indicators of post- approval drug safety and (2) to build and distribute preclinical indicator machine-learning models that predict post-approval drug safety. This proposal will deliver an open source drug safety sentinel that is based solely on preclinical data. The potential benefits to society include reduced exposure to unsafe drugs, an indicator for potentially suppressed safety data, and reduced burden on the FDA Adverse Event Reporting System and other phase IV surveillance systems. PUBLIC HEALTH RELEVANCE: The approval and subsequent withdrawal of widely prescribed unsafe drugs affects millions of lives annually. Patients affected by toxicity suffer from drug-induced morbidity and mortality, while those patients who benefited from the drug without toxicity can no longer receive it. We have recently reported that predictive models can predict efficacy and safety was accuracy. Our novel preliminary data demonstrate that predicting post-marketing safety issues independent of drug class is feasible using preclinical data only. Our objective is to build and distribute preclinical pharmacologic predictive models of post-approval clinical safety. Our goal is to deliver an open source drug safety sentinel that is based solely on preclinical data in order to reduce exposure to unsafe drugs.